Learning to rank search results
summary
Shows a concrete pipeline: index games in Elasticsearch, log clicks and basket clicks, generate a judgment list (expert or via epsilon click models), extract six features (title match, description, rating, release date, clicks, basket clicks), train models with RankLib/RankLib’s LambdaMART using ERR as metric, and deploy models into the Elasticsearch LTR plugin to rescore top 100 results. Demo improves some queries but warns of overfitting (some models hit 1.0) and needs more data and cleaning to be robust.
AI-written summary · about
chapters (28)
- 0:00Intro
- 5:53Elastic Search
- 10:38Learning to rank
- 16:25Example
- 18:37Recap
- 19:02Steps
- 19:30Judgement list
- 21:34Implicit Feedback
- 22:35Learning to rank plugin
- 22:56Expert panel
- 23:49Defining features
- 24:18Match on title
- 24:39Field Value Factor
- 24:58Underwater
- 25:14Click
- 25:25Logging features
- 26:47RankedLip
- 27:08Models
- 27:23Evaluation Metrics
- 28:28Deploy Model
- 29:27Testing
- 30:36Summary
- 31:19Click models
- 34:03Predicted judgement list
- 34:54Elasticsearch
- 35:04Perline
- 35:23Python Code
- 41:44Using more information
related talks
29mBuilding Learning to Rank models for search using LLMs
38mLearning to Rank journey at GetYourGuide: Our Logbook #bbuzz
55mLearning to Rank, a Little
31mSponsored Session: From Clicks to Rankings: Building a Learning-to-Rank Pipeline with…
26mTeaching Ruby to Rank: A Search Pipeline Story
43m[Machine Learning] Creating more relevant search results with Learn to Rank
38mLearning to Rank: where search meets machine learning #bbuzz
24mLearning "Learning to Rank"